The invention provides a method for encoding a progressive image based on adaptive block compressed sensing, belonging to the technical field of image ending. The method is characterized by comprising two encoding layers, namely a basic layer and a reinforced layer, wherein the basic layer is used for reading an image and dividing the image into blocks, observing the image blocks, sending observation values and observation rates to a decoding terminal so as to obtain an observation matrix according to the observation rates and obtain an initial solution according to the observation values and the observation matrix, carrying out Wiener filtering, updating with a picewise linear (PL) algorithm, carrying out contourlet wavelet transformation, carrying out bivariate shrinkage threshold processing on a wavelet coefficient, carrying out inverse wavelet transformation, updating with the PL algorithm, and repeatedly iterating until decoding is finished; and the reinforced layer is used for classifying the image blocks, observing the different classes of blocks with different observation rates, sending observation values and observation rates to the decoding terminal, using the restoration image of the basic layer as an initial value of the current decoding, carrying out the Wiener filtering, updating with the PL algorithm, carrying out contourlet wavelet transformation, carrying out bivariate shrinkage threshold processing, carrying out inverse wavelet transformation, updating with the PL algorithm, repeatedly iterating until the decoding is finished, and then carrying out observation and decoding of the next stage until the restoration image meets the requirements.